Harnessing Our Inner Angels and Demons

Page created by Victor Guerrero
 
CONTINUE READING
PE R SP EC TI V ES O N P SY CH O L O G I CA L S CIE N CE

Harnessing Our Inner Angels
and Demons
What We Have Learned About Want/Should Conflicts and
How That Knowledge Can Help Us Reduce Short-Sighted
Decision Making
Katherine L. Milkman, Todd Rogers, and Max H. Bazerman

Harvard University

ABSTRACT—Although       observers of human behavior have                    healthier in the long run rather than what would be more en-
long been aware that people regularly struggle with in-                     joyable in the short run and heads out the door in her exercise
ternal conflict when deciding whether to behave responsi-                   clothes. When her spouse wants to see a new action film and she
bly or indulge in impulsivity, psychologists and economists                 would prefer to watch the latest romantic comedy, despite the tug
did not begin to empirically investigate this type of want/                 she feels to behave selfishly, she compromises and does what
should conflict until recently. In this article, we review and              will give her spouse the most pleasure about half of the time
synthesize the latest research on want/should conflict, fo-                 (although he might debate the precise ratio).
cusing our attention on the findings from an empirical                         The types of internal conflicts described above are familiar to
literature on the topic that has blossomed over the last 15                 most people. In fact, the metaphor that individuals possess two
years. We then turn to a discussion of how individuals and                  selves—a want self fighting for whatever will bring more short-
policy makers can use what has been learned about want/                     term pleasure, and a should self representing an individual’s
should conflict to help decision makers select far-sighted                  long-term interests—is so common that its cinematic repre-
options.                                                                    sentation has become hackneyed. Films and television shows
                                                                            frequently depict internal conflict by showing a character with a
                                                                            whispering angel (the should self) perched near one ear and a
Recently, one of the authors of this article found herself strug-           fiery devil (the want self) at the other offering competing rec-
gling with the choice of what to order for dinner at her favorite           ommendations.
neighborhood Italian restaurant. With great difficulty, she de-                Evidence that storytellers have been aware of want/should
bated whether to indulge her craving for the restaurant’s sinfully          conflict for millennia can be found in Homer’s The Odyssey. In
delicious pizza or to stick to her diet and order a light salad. The        this epic tale, the hero Ulysses fears that, like many sailors
battle raged in her head until the waiter hovering over her                 before him, he will be lured by the desires of his want self to his
cleared his throat to signal that it was time for her to place an           death at the hands of the Sirens—sea nymphs whose sweet songs
order. In the end, she chose the pizza over the salad. The option           lead sailors to wreck their ships on the rocks surrounding the
she wanted more won out over the option she knew she should                 nymphs’ island. Before encountering the Sirens, Ulysses in-
select. However, we all face this type of internal conflict fre-            structs his crew to plug their ears and bind him tightly to his
quently, and we do not always succumb to our immediate desires              vessel so he will be able to listen to the Sirens’ song without the
at the expense of our long-term interests. Despite the impulse              power to turn his ship. In this way, Ulysses’ should self arranges
this author feels most afternoons to sit in front of the television         for the desires of his want self to be kept in check so disaster will
instead of going to the gym, she almost always does what is                 not befall him.
                                                                               Scholars have theorized for decades about various types of
                                                                            internal conflict people face (see, for example, Erikson, 1950/
Address correspondence to Max H. Bazerman, Baker Library, 453
Soldiers Field Road, Boston, MA 02163; e-mail: mbazerman@                   1963; Freud, 1923/1961; Higgins, 1997, 1998; Higgins, Roney,
hbs.edu.                                                                    Crowe, & Hymes, 1994; James, 1890/1990; Lecky, 1961;

324                                            Copyright r 2008 Association for Psychological Science                           Volume 3—Number 4
Katherine L. Milkman, Todd Rogers, and Max H. Bazerman

Markus & Nurius, 1987; Mead, 1934; Wylie, 1979). Much of this           COGNITIVE PROCESSES UNDERLYING WANT/SHOULD
literature has focused on the conflict people experience between                         CONFLICT
doing what is best for their long-term interests and what will
bring them the most immediate pleasure (Ainslie, 1975, 1992;            It is important to understand what cognitive processes underlie
Bazerman, Tenbrunsel, & Wade-Benzoni, 1998; Loewenstein,                the conflict people feel when deciding whether to base choices
1996; Schelling, 1984; Sen, 1977; Shefrin & Thaler, 1988; Strotz,       on what they want to do or what they feel they should do, because
1956; Thaler & Shefrin, 1981). Although there is a long history of      this knowledge will help us identify situational factors that are
theoretical research on this type of conflict, the last 15 years have   likely to affect the outcomes of intrapersonal conflicts, which, in
witnessed the publication of the first surge of empirical studies       turn, will help us determine how we can facilitate more should
examining the implications of what Bazerman et al. (1998) called        decisions. Bazerman et al. (1998) proposed that individuals
want/should conflict. In this article, we review the literature on      experience many decisions as if a want self and a should self
want/should conflict, with the majority of our attention focused        coexist within them and that these selves are susceptible to
on recent developments, and we discuss the implications of this         conflicting preferences. The want self is driven by the desires
stream of research for policy makers and individuals.                   people affectively feel in the moment when a decision will take
   We adopt the rational actor model from the field of economics        effect, whereas the should self is guided by more deliberative
as a backdrop for our work, a choice we believe is valuable for         feelings about what ought to be done given a person’s long-term
several reasons. First, and most importantly, it provides us with a     interests. A number of economic models have been proposed to
normative benchmark, allowing us to discuss what rational de-           explain intrapersonal conflict by assuming that people are ac-
cision makers ought to do in a given situation. The existence of a      tually controlled by multiple agents with different preferences
normative benchmark allows us to identify situations in which it        pitted against one another (Fudenberg & Levine, 2006; Read,
appears that people’s decisions systematically deviate from ra-         2001; Thaler & Shefrin, 1981).
tionality and thus permits us to discuss opportunities we see for          In contrast to these models of competing internal agents, other
improving people’s choices. Without a normative benchmark, we           research has focused on how different parts of the brain are
could still discuss opportunities for changing publicly lamented        triggered by different contexts. For example, recent brain
social behaviors like smoking, drug use, and undersaving for            imaging research (McClure, Laibson, Loewenstein, & Cohen,
retirement, but we could only subjectively assert that these            2004) has confirmed that two systems in the brain are involved in
behaviors are ‘‘problems.’’ Second, the framework provided by           decision making, one of which is preferentially activated by
economics allows us to articulate concrete definitions of the           decisions associated with the receipt of immediate rewards
concepts discussed in this article. Past research on want/should        (what Bazerman et al., 1998, would term the want self) and one of
conflict has been vulnerable to the criticism that the same be-         which is activated uniformly by decisions involving long-term
haviors could be classified, arbitrarily, as want or should options.    and short-term rewards (what Bazerman et al., 1998, would refer
Our model provides clear guidance on this question of classifi-         to as the should self). Loewenstein (1996) argued that intra-
cation. Finally, adopting an economic framework allows us to tie        personal conflicts stem from changes in the conditions under
together past work on self-control from the fields of economics         which decisions are made. He proposed that visceral factors,
and psychology. Linking research in these two areas enriches            such as emotions and psychological cravings like hunger, often
our understanding of the concepts of want and should.                   overwhelm people at the moment of a decision and that these
   In this article, we argue that the recent empirical literature on    visceral factors are the source of observed differences between
want/should conflict has identified ways in which individuals           the should preferences people often articulate when in a de-
and policy makers can design decision contexts in order to fa-          liberative state (‘‘I would like to lose weight’’) and the want
cilitate the selection of should options. To set the stage for our      preferences they often exhibit when making choices in a more
discussion, we summarize research on the cognitive processes            visceral state (‘‘I’ll take that cheeseburger with extra fries’’).
underlying want/should conflict and present a formal definition            Recent research on construal level theory (CLT) suggests that
of relative want and should options. We then review the latest          one fundamental difference between the want and should selves
empirical research on want/should conflict. After summarizing           is in the level at which they construe the world. CLT posits that
what is now understood about intrapersonal conflict and dis-            the temporal proximity of an event (and perhaps other factors as
cussing when want/should conflict most often leads to decision          well) systematically affects how it is construed (Liberman,
errors, we focus our attention on an exploration of how this new        Sagristano, & Trope, 2002; Trope & Liberman, 2003). Events in
knowledge can be applied. Specifically, we discuss how indi-            the distant future are construed at a high level, which means
viduals seeking to increase their chances of making should              they are associated with schematic, abstract, and purpose-
choices and policy makers hoping to improve the odds that               focused qualities, whereas events in the near future are con-
should policies will be adopted might be able to learn from             strued at a low level and are associated with concrete, specific,
studies of want/should conflict. We conclude with an assessment         detail-focused qualities. To give an example, a high-level
of opportunities for future research.                                   construal of a salad would focus on the salad’s healthfulness

Volume 3—Number 4                                                                                                                     325
Want/Should Conflicts and Short-Sighted Decision Making

and its likelihood of increasing longevity, whereas a low-level        likely be much healthier, slimmer, and happier if they manage to
construal of a salad would focus on its taste and its likelihood of    resist the temptation to eat pizza each night instead of salad.
leaving you hungry. A characteristic of choices that serve the         Now, think of films. An action-packed blockbuster is likely to be
interests of the should self (i.e., should choices) is that their      far more of a want film and less of a should film than a History
benefits are, essentially, future-goal directed and their costs are    Channel documentary is. Tying this back to our definition of
immediate and detail-focused. This leads to the prediction that        relative want and should options, a blockbuster is typically more
should choices are likely to be more attractive when construed at      exciting to watch than a history documentary. However, a history
a high level. This is what Rogers and Bazerman (2008) found            documentary is more likely to contain information that will serve
across eight different should choices, and Fujita, Trope, Liber-       you well in the future, enriching your conversations, helping you
man, and Levin-Sagi (2006) reported similar findings. Together,        impress your boss, and potentially even leading you to make
this research suggests that the interests of the should self are       better decisions. Finally, think of potential ballot issues. Voting
naturally aligned with higher level construal, whereas the in-         for a tax on gasoline is a should behavior (it will help reduce
terests of the want self are naturally aligned with lower level        emissions, thereby improving air quality and the environment in
construal.                                                             the future), but voting against it is something that many citizens
                                                                       want to do (‘‘I love filling up my gas guzzling SUV on the
  A FORMAL DEFINITION OF RELATIVE WANT AND                             cheap!’’). Voting for the should policy is likely to bring you more
              SHOULD OPTIONS                                           happiness in the future because it will increase your chances of
                                                                       breathing cleaner air and slowing global warming. It will also
To consider the applications of research on want/should conflict,      allow you to feel pride whenever you recall your vote because
we must also have an understanding of what types of options lead       you will know that you contributed to a good cause. However,
individuals to experience such conflict when making decisions.         voting for the want policy will increase the odds that you will be
We propose a definition for options that are associated with the       able to enjoy your SUV in the short run.
preferences of the want and should selves. Certain options are
naturally preferred by the should self (e.g., salads, documentary       RECENT EMPIRICAL RESEARCH ON WANT/SHOULD
films, trips to the gym), whereas others are naturally preferred by                     CONFLICT
the want self (e.g., ice cream cones, action films, skipping the
gym). Given two options, we define one option as having rela-          Recent empirical research on want/should conflict has primarily
tively more want and fewer should characteristics than a second        focused on identifying situational factors that cause individuals
option if and only if the following two conditions hold:               to reverse their preferences for want options over should options.
                                                                       In this section, we review the literature on different conditions
1. The instantaneous utility obtained from the want option is          known to induce want/should preference reversals. The knowl-
   greater than the instantaneous utility obtained from the            edge gained from the empirical studies we discuss in this section
   should option.                                                      will be the cornerstone of our upcoming discussion of how in-
2. The sum of the utility (discounted at a standard rate, d51 e)       dividuals and policy makers may be able to design decision-
   that will be derived from the want option in all future periods     making contexts that facilitate more should choices and why it
   is less than the sum of the utility that will be derived from the   makes sense for them to do so.
   should option in all future periods.

It is important to note that this definition does not classify         Intertemporal Choice
whether a want or should option is optimal. The optimal choice         The majority of past research on want/should conflict has been
between want and should options requires summing the short-            focused on the domain of intertemporal choice. To summarize, it
run and long-run utility that would be gained from each option         has been hypothesized and confirmed in a variety of contexts
and selecting whichever provides more discounted net utility.          that people are considerably more likely to favor should options
Although should options have more long-run benefits than want          over want options when making choices that will take effect in
options, in many cases the short-run benefits of a want option         the future than they are when making decisions that will take
may be significant enough to outweigh the long-run benefits of a       effect immediately. This observation holds when decisions are
should option.                                                         made in the domains of money (Ainslie & Haendel, 1983; King
   To put our definition of want and should options in context,        & Logue, 1987; Kirby, 1997; Kirby & Herrnstein, 1995; Kirby &
consider some examples. First, think of foods. According to our        Marakovic, 1996; McClure et al., 2004; Thaler, 1981), exercise
definition of things associated with the want and should selves,       (Della Vigna & Malmendier, 2006), and film choice (Milkman,
pizza is a want good, whereas salad is a should good. Pizza gives      Rogers, & Bazerman, 2008a; Read, Loewenstein, & Kalyanara-
most consumers more instant gratification than salad while it is       man, 1999), among others. For example, to most people, the idea
being consumed (yum, grease!). However, the future utility             of going to the gym tomorrow is much more palatable than the
gained from eating the salad is higher because consumers will          idea of going to the gym this minute. Similarly, the idea of

326                                                                                                                      Volume 3—Number 4
Katherine L. Milkman, Todd Rogers, and Max H. Bazerman

starting a diet next week sounds much more appealing than the                Now that we have discussed the theoretical work on dynamic
idea of starting a diet today. In this subsection, we will briefly        inconsistency, we move to a discussion of the empirical research
discuss how want/should conflict has been modeled in the                  that has been conducted to test the prevailing theory. Before
context of intertemporal choice before reviewing the empirical            people were using terms like should and want to describe
research in this area.                                                    multiple-selves conflict, a number of laboratory studies were
   Researchers began modeling dynamically inconsistent pref-              conducted to examine impulsiveness in the domain of money. In
erences as early as 1956 with nonstandard time discount func-             an early study of dynamic inconsistency, Ainslie and Haendel
tions (Strotz, 1956). To fit the observation that people have great       (1983) conducted a laboratory experiment in which subjects
difficulty passing up a large reward in the present for a larger          were divided into two conditions. In one condition, they were
reward tomorrow and considerably less difficulty passing up the           given a hypothetical choice between receiving $50 immediately
same large reward tomorrow for that larger reward in 2 days,              (the want option) or $100 in 6 months (the should option), and in
economists have modeled individuals as possessing a discount              the second condition they were given a hypothetical choice
rate for utility that is extremely high in the short run but rela-        between receiving $50 in 12 months or $100 in 18 months.
tively low in the long run (Ainslie, 1992; Laibson, 1996;                 Standard economic theory suggests that the choices made by
Loewenstein & Prelec, 1992; Strotz, 1956). A leading model of             subjects in the two conditions should be indistinguishable.
impulsiveness in the economic literature is Laibson’s (1996)              However, Ainslie and Haendel found that most subjects chose
quasi-hyperbolic time discount model, which models the ex-                the smaller reward in the first condition, when it would be
treme short-run drop in valuation that has been observed in               available immediately, and the larger reward in the second
people’s time preferences by adding a discount factor of b,               condition, when all rewards would be delayed. A plethora of
which is much less than 1, to all but the first time period of a          other field and laboratory studies have been conducted since
traditional discrete-time exponential discount model. Specifi-            then, using real monetary payoffs as well as hypothetical pay-
cally, Laibson’s model assumes that individuals place no dis-             offs, and all have confirmed that people exhibit an abnormally
count on immediate utility, but that they discount all future             high discount rate between immediate and delayed payoffs
utilities by b in addition to the traditional, exponential discount       (Angeletos, Laibson, Repetto, Tobacman, & Weinberg, 2001;
rate, d, which is much greater than b. For partial reviews                Kirby, 1997; Kirby & Herrnstein, 1996; Kirby & Marakovic,
of the economic literature on dynamic inconsistency, see                  1996; McClure et al., 2004). However, many of these studies
Ainslie (1992), Frederick, Loewenstein, and O’Donoghue                    have been subject to a number of criticisms. First, if subjects
(2002), Loewenstein and Thaler (1989), and O’Donoghue and                 believe there is some risk that an experimenter will not follow
Rabin (1999).                                                             through on payoffs offered in the future, measured short-term
   Combining Laibson’s (1996) model with the definition of                discount rates may be biased upward. Second, if utility flows
should and want options discussed earlier produces the pre-               result from the things that money can buy rather than the money
diction that people will often reverse their preferences over             itself, studies conducted with monetary payoffs should not in-
relative want and should items like an action film and a docu-            duce as much short-term utility as would studies conducted with
mentary when the delay between selection and consumption                  real goods as payoffs, so measured short-term discount rates
switches between short and long. For example, consider two                might be biased downward.
movies that you could rent for tomorrow night: a want film                   Empirical studies of the multiple-selves phenomenon focus-
(Steven Spielberg’s action blockbuster ‘‘Jurassic Park’’) or a            ing on choices between relative should and want options in
should film (Steven Spielberg’s Academy Award winning film                realms besides money followed research on monetary discount
about the Holocaust ‘‘Schindler’s List’’).’’ If ‘‘Schindler’s List’’ is   rates. Such studies allowed academics to confirm that previously
a should film relative to ‘‘Jurassic Park’’ and if it is also your        observed anomalies of intertemporal choice were not caused by
optimal film choice because it provides more net utility, you will        anything special about the way people value money relative to
rent ‘‘Schindler’s List’’ when choosing which film to watch to-           other goods. One nonmonetary domain where multiple studies of
morrow. This is because the utility flows from both movies will be        want/should conflict have been conducted is that of film rentals.
uniformly downweighted by b when you consider which film to               Read et al. (1999) conducted an experiment in which subjects
watch tomorrow, as all utility flows from each movie will be              were given an opportunity to rent three films for three future
achieved in the future.                                                   dates from a list of 14 highbrow (should) and 10 lowbrow (want)
   However, when tomorrow arrives and you are faced with the              movies. Study participants were randomly assigned to one of two
decision of which film to watch today, you may choose to watch            experimental conditions. In the first condition, they were asked
‘‘Jurassic Park’’ instead of ‘‘Schindler’s List’’ as a result of the      to simultaneously choose 3 films from the list of 24 that they
high discount (b) you place on future utility flows and the fact          would like to rent in the future. In the second condition they
that the immediate rewards from watching ‘‘Jurassic Park’’ ex-            were allowed to choose the 3 films sequentially on the days when
ceed the immediate rewards from watching ‘‘Schindler’s List’’             their rentals would take place. The authors found that subjects
(by the definition of relative want and should options).                  in the sequential choice condition, who were able to select films

Volume 3—Number 4                                                                                                                      327
Want/Should Conflicts and Short-Sighted Decision Making

right before consumption rather than well in advance of con-          spent more when ordering for more immediate delivery
sumption, rented significantly more lowbrow movies and fewer          (spending is a typical want behavior, whereas saving is a should
highbrow movies than did subjects in the simultaneous choice          behavior), Milkman et al. (2008b) determined that the per-
condition. This finding is consistent with the prediction of the      centage of extreme should groceries in a customer’s basket
theories presented above about intertemporal choice and               generally increases the further in advance of delivery an order is
should/want conflict. When making decisions for the present,          completed, whereas the percentage of extreme want groceries in
subjects in this study were more susceptible to the whims of their    a customer’s basket generally decreases the further in advance
want selves, but when choosing for the future, subjects were          of delivery an order is completed.
more likely to rent should movies.                                       Related research has been conducted by Oster and Scott
   Milkman et al. (2008a) followed up on this research by Read        Morton (2005) on the newsstand and subscription prices for
et al. (1999) with a field study of dynamic inconsistency in the      leisure magazines (want magazines), or magazines that are fun to
domain of online DVD rentals. In order to study intertemporal         read ‘‘now,’’ and investment magazines (should magazines), or
choice and want/should conflict among online DVD rental               magazines that provide benefits in the future. These authors
customers, the authors obtained 4 months of data on individual        recognized that if the type of internal want/should conflict dis-
customers’ rental and return decisions from an Australian online      cussed in this article has a significant effect on decision making,
DVD rental company. Milkman et al. (2008a) began by classi-           magazine prices should reflect the fact that people will rarely
fying the thousands of films offered for rent by this company         plan ahead when it comes to the consumption of leisure maga-
along a continuous scale from extreme should films to extreme         zines (e.g., People) but will regularly plan ahead when it comes
want films. To classify the films, the authors used a survey in       to the consumption of investment magazines (e.g., The Econo-
which respondents were asked to rate the extent to which 500          mist). In an efficient market, such behavior should lead the ratio
different films were should and want movies. The films’ average       of a magazine’s newsstand price to its subscription price to be
ratings were then used to estimate a regression equation for          considerably larger for a leisure magazine than for an investment
predicting a film’s average should-minus-want score based on its      magazine. Oster and Scott Morton find that this is the case in
quantifiable characteristics (e.g., genre, year of release, average   their study of approximately 300 American magazines.
user rating), and this equation was used to give all films in the        Della Vigna and Malmendier (2006) conducted a study that
authors’ database should-minus-want scores. After creating this       examined gym attendance (a should behavior) and found evi-
scoring system, Milkman et al. (2008a) tested and confirmed the       dence that people exhibit dynamic inconsistency in the domain
hypothesis that people are more likely to rent DVDs in one order      of exercise. The authors analyzed data on the gym contracts
and return them in the reverse order when should DVDs (e.g.,          purchased by thousands of gym customers, as well as the gym
documentaries) are rented before want DVDs (e.g., action films).      attendance records of those customers. They found that people
In addition, the authors predicted and found that should DVDs         regularly paid a high fee for the right to visit their gym an un-
are held longer by customers than want DVDs. Although these           limited number of times when they could have saved money by
field results were consistent with previous theory and laboratory     paying flat per-visit fees instead. These findings are consistent
research on intertemporal choice, they are notable because they       with a model of consumers who make should choices when
confirmed that the effects of want/should conflict are large en-      thinking about the future (i.e., they pay to go to the gym fre-
ough to significantly affect real-world decision making. Perhaps      quently in the future) but systematically reverse their prefer-
more noteworthy still, Milkman et al. (2008a) determined that a       ences and opt for want options when the future eventually
movie’s position on the want/should spectrum is as strong a           becomes the present (i.e., they do not actually go to the gym
predictor of how long that movie will be held by an online DVD        frequently).
rental customer as are all of the film’s other quantifiable char-
acteristics (e.g., genre, year of release, average user rating)
combined.                                                             Joint Versus Separate Decision Making
   Another domain in which recent research has been conducted         Want/should conflict is also evident in reversals of preference
on want/should conflict and intertemporal choice is the domain        that have been observed in joint versus separate decision
of groceries. Milkman, Rogers, and Bazerman (2008b) examined          making. Academics have hypothesized and confirmed in a va-
dynamic inconsistency in this context by obtaining a year of data     riety of settings that individuals are more likely to favor want
detailing the orders placed by the customers of a major North         options over should options when evaluating different possibil-
American online grocer. The authors examined how the mix of           ities one at a time rather than simultaneously (see Bazerman
should and want goods purchased by the same shoppers differed         et al., 1998, and Bazerman, Moore, Tenbrunsel, Wade-Benzoni,
depending on how far in advance of delivery an order was              & Blount, 1999, for reviews of this literature). For example, the
completed. Goods were assigned should and want scores on the          idea of donating to a charity that protects baby polar bears
basis of the average score survey respondents assigned to gro-        (a relative want option) may generate more enthusiasm than the
ceries in their category. In addition to finding that customers       idea of donating to a charity that supports research on malaria

328                                                                                                                     Volume 3—Number 4
Katherine L. Milkman, Todd Rogers, and Max H. Bazerman

(a relative should option) when each donation opportunity is           where they would be paid as much as other new hires, over the
considered separately. However, when given a choice between            should job, where they would earn more overall but less than
saving baby polar bears and reducing the numbers of human              other new hires.
deaths from malaria, most people feel more obliged to save                Shortly after Bazerman et al.’s (1992) initial study of joint
people than polar bears (Kahneman & Ritov, 1994). Bazerman             versus separate preference reversals was published, Kahneman
et al. (1998) proposed that should options are more likely to win      and Ritov (1994) conducted a similar study of individuals’
out over want options in joint evaluation than they are in sep-        willingness to give to different types of charities when donation
arate evaluation because direct comparison forces an individual        opportunities were presented jointly versus separately. Kahne-
to rationally weigh the costs and benefits of her choices. The         man and Ritov (1994) identified a pattern of preference reversals
short-term, visceral desires of the want self are more likely to       between joint and separate evaluations similar to those observed
overwhelm a decision maker in separate evaluation than in joint        by Bazerman et al. (1992) and Bazerman et al. (1994). In their
evaluation because there is no explicit tradeoff to cue the            study of people’s willingness to give, subjects were presented
rational weighing of costs and benefits in separate evaluation.        with opportunities to donate to wildlife charities (e.g., a charity
   In the first explicit study of preference reversals between joint   that protected kangaroos) or charities supporting human health
and separate evaluation, Bazerman, Loewenstein, and White              or safety (e.g., a charity that provided free skin cancer check-ups
(1992) evaluated the likelihood that participants in a fictional       for farm workers). The wildlife charities fit the definition of a
dispute with a neighbor would accept two different settlement          want option because they are more immediately emotionally
options depending on whether those options were presented              attractive, whereas the charities promoting human health and
jointly or separately. In one settlement option, both disputants       safety are more pragmatically important in the long run by most
would receive the same amount of money. In the second settle-          measures and thus fit the definition of a should option. In joint
ment option, both disputants would receive more money than             evaluation, the should options were systematically preferred
they would have under the terms of the first settlement, but the       over the want options by study participants, but in separate
participant’s neighbor would receive a larger payout than the          evaluation, this pattern reversed itself as in the Bazerman et al.
participant. The second option was considered the should option        (1992) and Bazerman et al. (1994) studies.
because it yielded a higher payoff to the participant and would           In another early study of this phenomenon, Irwin, Slovic,
thus provide more long-term utility, whereas the first option was      Lichtenstein, and McCelland (1993) compared people’s will-
considered the want option because the participant would not           ingness to pay for public goods (e.g., air quality improvements)
have to experience the short-run, visceral displeasure associ-         with their willingness to pay for commodities for their personal
ated with receiving less than her neighbor. In the separate            use (e.g., a new VCR) in both joint and separate evaluation. In a
evaluation condition, the two settlements were presented to            study whose findings were consistent with those of others, Irwin
participants sequentially, and participants were asked to rate         et al. (1993) determined that in joint comparison, public goods
the appeal of each option on a continuous scale. Under these           (or should goods) elicited a higher willingness to pay from study
circumstances, the average participant rated the option in which       subjects than did commodities for personal use (or want goods),
she and her neighbor would receive equal payoffs (the want             but they found that this pattern reversed itself in separate
option) more favorably than the option in which she would re-          evaluation.
ceive less than her neighbor but would receive more money                 Lowenthal (1993) conducted a study of joint versus separate
overall (the should option). However, in the joint evaluation          preference reversals that examined people’s willingness to
condition, which pitted the two settlement options directly            vote for different candidates running for political office. In
against one another, the average participant preferred the set-        Lowenthal’s study, a candidate boasting the ability to bring
tlement in which she would receive a higher payoff (the should         1,000 new jobs to his district and a clean criminal record (the
option).                                                               want candidate) was pitted against a candidate who had been
   A subsequent study by Bazerman, Schroth, Shah, Diekmann,            convicted of a misdemeanor but could produce 5,000 new jobs
and Tenbrunsel (1994) replicated the basic preference reversal         for his constituents (the should candidate). When evaluated
result described above. In this study, participants were asked to      jointly in an ‘‘election,’’ the should candidate earned more votes,
consider accepting matched hypothetical job offers. In a typical       but when rated separately on likeability, the want candidate was
comparison, one job, the want job, would pay the participant and       more popular. These studies offer yet more demonstrations of
other new hires the same yearly salary. The other job, the should      want/should preference reversals between joint and separate
job, would pay the participant a higher yearly wage than the first     evaluations.
job but less than it would pay other new hires. When evaluated            Tenbrunsal, Wade-Benzoni, O’Connor, and Bazerman (1997)
jointly, the vast majority of participants expressed a preference      sought to confirm that the want/should framework accurately
for the should job over the want job, presumably due to its su-        characterizes the types of joint versus separate preference re-
perior paycheck. When evaluated separately, however, a higher          versals described above (see Bazerman et al., 1998, for a more
percentage of participants said they would accept the want job,        detailed description of this work). Tenbrunsal et al. showed

Volume 3—Number 4                                                                                                                    329
Want/Should Conflicts and Short-Sighted Decision Making

subjects similar and, in some cases, identical materials to those       the outcomes of choices between should and want options in
used in the Bazerman et al. (1994), Kahneman and Ritov (1994),          contexts besides those previously discussed, such as extreme
and Irwin et al. (1993) studies, as well as several other similar       cognitive load (Shiv & Fedorkin, 1999), whether an individual
studies of joint versus separate decision making. They then             views a choice as isolated or as the first in a sequence of related
asked subjects which option from each study a typical student           choices (Khan & Dhar, 2007), and whether an individual feels
would want to choose and which option they would normatively            licensed to make want choices as a result of recent should be-
feel they should choose. Across contexts, students identified the       haviors (Khan & Dhar, 2006).
want option as the option that had been preferred in separate              Shiv and Fedorikhin (1999) conducted a study to examine the
evaluation and the should option as the option that had been            impact of cognitive load on people’s preferences for foods as-
preferred in joint evaluation significantly more often than not.        sociated with intense positive emotions (want foods) versus
Tenbrunsal et al.’s findings lend strong support to our contention      foods associated with less positive affect but more favorable
that want/should conflict is consistent with the preference re-         cognitions (should foods). Participants in their study were ran-
versals described above.                                                domly assigned to a high or low cognitive load condition. In the
   Hsee (1995, 1996, 1998) examined whether joint versus                high cognitive load condition, subjects were instructed to
separate preference reversals might be driven by what he termed         memorize a seven digit number, and in the low cognitive load
the evaluability hypothesis. Hsee’s evaluability hypothesis pro-        condition they were asked to memorize a two digit number. Study
poses that when people evaluate options with multiple attributes        subjects were then given a choice between two snacks: a slice of
in joint evaluation, it will be possible for them to compare and        chocolate cake (a want snack) and a cup of fruit salad (a should
weigh each attribute appropriately, but when they evaluate using        snack). Shiv and Fedorikhin hypothesized that subjects in the
separate evaluation, the attributes with clearer standards for          high cognitive load condition would be more likely to select the
evaluation (e.g., GPA, SAT scores) will be overweighed relative         chocolate cake than would subjects in the low cognitive load
to attributes with less clear standards for evaluation (e.g., a score   condition because fewer of their intellectual resources would be
on an unknown test), resulting in preference reversals between          available to help them resist temptation. The authors found that
joint and separate evaluations. Hsee conducted a number of              significantly more subjects selected cake over fruit salad in the
studies to test his evaluability hypothesis. In one study, he           high cognitive load condition (63%) than in the low cognitive
demonstrated that people are more likely to prefer a dictionary,        load condition (42%).
A, with a torn cover and twice as many entries as a second                 Khan and Dhar (2007) evaluate the difference in choices
dictionary, B, with a immaculate cover when the two are com-            people make between want and should goods when making a
pared jointly, but he found that this preference reverses itself        one-shot decision versus the first in a series of similar decisions.
when the dictionaries are presented separately (Hsee, 1996).            The authors conducted three studies in which subjects
Similarly, Hsee (1998) found that people using joint evaluation         were given the opportunity to choose between an array of virtue
would prefer a 40-piece set of china with 9 broken pieces to an         (should) and vice (want) goods. Subjects in all three studies were
undamaged set with just 24 pieces, but people using separate            randomly assigned to one of two conditions and given the op-
evaluation find the unmarred china set more appealing. Hsee             portunity to select one magazine, one movie, or one snack for
conducted similar studies with ice cream cups, CD changers,             immediate enjoyment. In the first treatment condition, subjects
and job applicants, among other things. However, although the           were told that their choice would be the first of a series of similar
pattern of preference reversals Hsee identified is consistent with      choices, the rest of which would be made in the future. In the
his evaluability hypothesis and though Hsee’s hypothesis may            second condition, subjects were told they would be making
offer the best explanation for these reversals, it is also consistent   an isolated choice. Subjects selected a significantly higher
with the idea that want items are more often preferred over             proportion of virtuous movies, magazines, and snacks when
should items in separate evaluation than in joint evaluation. A         they believed their choice was made in isolation and was
damaged dictionary or china set creates a negative visceral re-         not the first of a series of similar choices. Khan and Dhar
action, so an undamaged good with fewer of the qualities we             (2007) hypothesized that subjects gave into their visceral de-
know we ought to care about is a want option, whereas a dam-            sires more frequently when they believed they were making the
aged good is a should option.                                           first in a series of similar choices because they were able to
                                                                        offset the guilt associated with their impulsive behavior by
Other Moderators of Want/Should Conflict                                anticipating that their future selves would make more virtuous
Although the majority of empirical research on want/should              selections. In a study whose results were consistent with this
conflict to date has focused on intertemporal choice and joint          hypothesis, the authors found that when subjects in the re-
versus separate decision making, these are not the only contexts        peated choice condition of their snack experiment believed
in which differences in a decision maker’s situation can sys-           they would be eating a virtuous snack the following week, the
tematically sway the desires of her want self or the pragmatism of      percentage choosing a virtuous snack for immediate con-
her should self. Recent studies have examined forces that affect        sumption decreased.

330                                                                                                                         Volume 3—Number 4
Katherine L. Milkman, Todd Rogers, and Max H. Bazerman

   In a related study, Khan and Dhar (2006) examine differences      strategies that will help them and their constituents, respec-
in choices people make between want and should goods when            tively, increase their odds of choosing should options over want
they make their decision after engaging in a should behavior in a    options.
separate domain. They hypothesize that engaging in a should             By definition, when the difference between the net future
behavior makes people feel ‘‘licensed’’ to make want choices in      utility of a should option and a want option is greater than the
the near future, a hypothesis supported by studies conducted by      difference between the immediate utility those want and should
Monin and Miller (2001) showing that people feel more licensed       options have to offer, the should option is the optimal choice. We
to exhibit prejudice after establishing themselves as unpreju-       argue that the majority of decision-making errors resulting from
diced. In their first study, Khan and Dhar (2006) find that sub-     want/should conflict arise when a should option is optimal but a
jects in a hypothetical forced-choice task are significantly more    want option is selected instead. Angeletos et al. (2001) provided
likely to choose to buy a pair of designer jeans (the want choice)   evidence that this is the case with a study that estimated
than a vacuum cleaner (the should choice) after imagining            Americans’ average discount rates based on retirement wealth
spending 3 hr per week volunteering for community service (a         data. According to their calibration, between the present and
should behavior). In their second study, the authors find that       one year in the future, Americans discount money at a rate of .53,
subjects are more likely to report that they would spend a tax       meaning that on average, people view $1.00 in one year and
rebate check on a pair of expensive, designer sunglasses (the        $0.53 today as equivalent. It is difficult to argue that such a steep
want choice) than on a pair of less-expensive, utilitarian sun-      discount rate is optimal in an environment where yearly inflation
glasses (the should choice) if they have just imagined donating      has averaged 2.7% over the last decade.1 Indeed, Angeletos
$100 from their tax rebate check to a charity. The authors also      et al. reported that 55% of respondents in a 1997 survey said
show that subjects who are asked to indicate if they would help a    they were behind on their savings goals, whereas only 6% re-
foreign student who requested assistance understanding a lec-        ported being ahead. The fact that America’s obesity problem is
ture (a should behavior) donate less to charity (a want behavior)    so extreme (66% of Americans are overweight or obese; see
than do students who are not asked to indicate if they would help    Medline Plus, 2007) also suggests that people more frequently
a foreign student. In addition to establishing this licensing ef-    suboptimally overweight the desires of their want selves relative
fect, Khan and Dhar show that if participants are told to imagine    to those of their should selves than visa versa. The preponder-
that they engaged in a should behavior involuntarily, they are no    ance of attention given to self-control problems in the psy-
more likely to make a want choice than they would be in a            chology and economics literatures compared with that given to
baseline condition. Thus, the authors conclude that people are       underindulgence problems (see the Appendix for a comparison)
more likely to make want choices and less likely to make should      also suggests that self-control problems are more common than
choices after voluntarily engaging in a should behavior, but not     underindulgence problems. Evidence from a survey we con-
after being forced to engage in a should behavior.                   ducted with a national sample of paid participants also supports
                                                                     this view, although there are alternative explanations for its
                MAKING OPTIMAL CHOICES                               results. We found that 66% of respondents reported making
                                                                     errors that involved favoring their want selves over their should
Before turning to a discussion of ways in which the findings from    selves when it was suboptimal to do so more frequently than
the literature on want/should conflict can be applied to help        mistakenly favoring their should selves over their want selves.
individuals and policy makers, it is important to address the           In addition to being the more common error, the mistake of
question of when want/should conflict leads to decision-making       suboptimally undervaluing should options relative to want op-
errors. As discussed previously, given a choice between a want       tions is an error that can often lead to far more detrimental
option and a should option, a perfectly rational decision maker      consequences than the opposite mistake. For instance, consider
would select whichever option will yield more exponentially          the following potential outcomes of self-control problems: un-
discounted net utility. When a small change to the context in        dersaving for retirement in order to enjoy a more indulgent
which a decision is being made systematically leads to pre-          lifestyle while in the work force, becoming a drug addict or an
dictable reversals in people’s preferences for should options        alcoholic in order to enjoy popularity and the thrills associated
versus want options, it is safe to assume that one context is in-    with drugs and alcohol, failing to earn a high school diploma in
ducing a decision error, as the same choice typically remains        order to participate in more extracurricular activities, and
optimal across the contexts. In this section, we argue that de-      paying late fees. Now consider the outcomes of underindulgence
cision errors that involve favoring want options when should         problems in the same domains: oversaving for retirement at the
options are optimal occur more frequently and are more detri-        expense of a more indulgent lifestyle while in the work force,
mental than errors that involve favoring should options when         missing out on the thrills and popularity frequently associated
want options are optimal, although there is evidence that both
types of mistakes occur. As a result, we propose that individuals      1
                                                                         As computed by the Consumer Price Index Inflation Calculator (http://data.
and policy makers should focus their attention on developing         bls.gov/cgi-bin/cpicalc.pl, accessed on September 13, 2007).

Volume 3—Number 4                                                                                                                             331
Want/Should Conflicts and Short-Sighted Decision Making

with drug and alcohol use, receiving ‘‘excess’’ education and         tions in which these choices are likely to be suboptimal, can
spending too little time on more enjoyable extracurricular pur-       help us think about strategies for solving important problems
suits, and paying bills so early that some interest is foregone.      that result from flawed decisions. With a better understanding of
These examples suggest that errors in judgment that stem from         the conditions that affect whether people lean towards want or
overweighting want options relative to should options often have      should options, individual decision makers may be able to help
far more severe consequences than do errors caused by over-           themselves make more choices that are optimal but that con-
weighting should options relative to want options, which is an-       tradict what they want to do, and policy makers may also be able
other reason why we believe policy makers and individuals             to help facilitate more optimal should decisions. In this section,
interested in reducing the negative effects of suboptimal deci-       we describe recent empirical research demonstrating how peo-
sion making should focus on finding strategies to increase the        ple can and often do take steps to increase the likelihood that
odds that people will make should choices.                            they will follow the advice of their should selves. This work
   Thus far, we have focused our attention on the most common         suggests that people are eager to find ways to better manage their
mistake associated with want/should conflict: the error that          intrapersonal conflicts and demonstrates how some individuals
leads people to overweight the desires of their want selves.          have effectively managed to help themselves make more should
However, as discussed above, sometimes people make the op-            choices. We also discuss empirical work with implications for
posite error. That is to say, they underweight their short-term,      how policy makers may be able to design decision contexts that
want desires and do what will provide more future utility, even       facilitate the selection of should options.
when doing so is suboptimal. Assuming people aim to maximize             It is important to note that in this section we do not discuss
their utility, it is optimal to choose a want option when the im-     ways in which incentives can be manipulated to change the
mediate utility from that option exceeds the immediate utility        likelihood that people will select should options. Many policies
from a should option by more than the should option’s net future      that favor should options by changing incentives have been
utility exceeds that of the want option. However, sometimes           proposed and implemented. For example, placing ‘‘sin’’ taxes on
people make should choices when a want choice would make              cigarettes and alcohol and outlawing heroin are ways of in-
them better off overall: for example, abstaining from enjoyable       creasing the likelihood that people will make should choices by
pleasures like alcohol and donuts when they are in good health        raising the costs associated with want choices. Although such
or oversaving for retirement instead of buying a nicer house.         policies may have many benefits (see Gruber & Köszegi, 2004,
   Kivetz and Simonson (2002b) have demonstrated that people          for example), we focus our attention on discussing ways that the
are aware that they sometimes choose should options when they         should self can be given a leg up without directly manipulating
would be better off choosing want options. In a series of studies,    the costs associated with want options. Unlike incentive ma-
they found that a large proportion of people are willing to pre-      nipulations, the methods we discuss for increasing the likeli-
commit to future want options that are objectively inferior to        hood that people will make should decisions do not require
available should options. For example, Kivetz and Simonson            policy makers to decide what choices are in everyone’s long-run
(2002b) found that 28% of subjects waiting in an airport would        best interest. Instead, we examine the conditions that will help
choose a bottle of wine valued at $50 (the want option) as a prize    each individual do what is in her own long-term best interest.
over $55 in cash (the should option). They argue that this is evi-    The choices of individuals who do not face want/should conflict
dence that some people are willing to ‘‘precommit to indulgence’’     in a given domain will not be altered by the methods we propose
because they know they will suboptimally underindulge otherwise.      for increasing the incidence of should decision making, whereas
   Although there is evidence that people occasionally make the       everyone who drinks alcohol would be affected by an increase in
suboptimal decision to underindulge in want options, it seems         its price, regardless of whether drinking less alcohol represents
that people are considerably more likely to overindulge in want       a should choice for them.
options. In addition, we have argued that the penalties associated
with overindulgence in want options are larger than those asso-
ciated with underindulgence in want options. For these reasons        Commitment Devices
we focus our discussion of the implications of research on want/      Some researchers interested in intertemporal want/should con-
should conflict on how decision makers and policy makers can          flict have begun to explore the question of whether people are
increase the rate at which they and their constituents, respec-       willing to take measures to prevent themselves from acting on
tively, select should options when experiencing internal conflict.    want impulses instead of doing what they feel they should.
                                                                      Preventative measures taken to restrain the want self are fre-
  APPLICATIONS OF RESEARCH ON WANT/SHOULD                             quently referred to as commitment devices, which typically re-
                  CONFLICT                                            quire an individual to commit to making a should choice in the
                                                                      present rather than a want choice in the future. Some examples
Understanding the conditions in which people select should            of commitment devices that many people are familiar with in-
options over want options and vice versa, as well as the condi-       clude piggy banks, which people’s should selves use to prevent

332                                                                                                                    Volume 3—Number 4
Katherine L. Milkman, Todd Rogers, and Max H. Bazerman

their want selves from dipping into their savings; diet treatment      that would prevent them from doing what they wanted to do
centers, which people’s should selves enroll in to prevent their       rather than what they felt they should.
want selves from overeating; and pills like Antabuse, which               Ashraf, Karlan, and Yin (2006) conducted a study of people’s
people’s should selves take so their want selves will face the         willingness to take up commitment devices in the domain of
prospect of a painful illness if they consume an alcoholic beverage.   savings. The authors partnered with a bank in the Philippines,
   In an early study of commitment devices, Wertenbroch (1998)         which contacted a group of its former clients to offer them a
used supermarket scanner data to conduct a paired comparison           choice between two savings products. The first savings product
of the quantity discounts applied to a matched sample of 30            was a standard savings account, but the second was a commit-
virtue (should) and 30 vice (want) grocery products. He found          ment savings account from which money could not be withdrawn
that, on average, vice foods are subject to steeper quantity dis-      until the depositor reached a self-selected, predetermined date
counts than virtue foods are, and the demand for virtue goods is       or savings goal. Money in both accounts earned the same rate of
less price sensitive than the demand for vice goods is. These two      interest, and thus people only had an incentive to place their
findings suggest that consumers are aware of their impulsivity         money in the commitment savings account if they wanted to
and that their should selves take steps at the time of purchase        prevent their want selves from impulsively withdrawing and
(which is in advance of consumption) to prevent their want             spending funds set aside for the future by their should selves.
selves from having the opportunity to binge in the future. In          The study by Ashraf et al. (2006) had two important findings.
other words, people are willing to pay more to buy smaller             The first was that a significant percentage of people (28%) were
packages of vice foods to avoid having too many such foods             willing to give up the freedom to withdraw money from a savings
around to tempt their impulsive, want selves when they sit down        account at their discretion without accepting any compensation
to eat. Buying small packages of want foods helps people commit        for this sacrifice, indicating that they placed a positive value on
to eating less junk food than they might otherwise, and people         the commitment device. The second finding was that individuals
are willing to pay a price for this commitment device.                 who were given the opportunity to use a commitment savings
   One interpretation of the study of gym contracts and gym at-        product saved 81% more over the course of a year than indi-
tendance conducted by Della Vigna and Malmendier in 2006,              viduals who were not offered this product, indicating that people
which we discussed previously, is that it provides evidence that       are able to save more when their want selves are kept in closer
people place a positive value on commitment devices. Della             check. These findings suggest that those who understand the
Vigna and Malmendier found that people often pay for gym               implications of dynamic inconsistency may be able to use their
memberships that entitle them to unlimited gym visits despite          knowledge to design policies with the potential to benefit society
the fact that they would save money if they simply paid                by increasing savings rates, among other things.
per-visit usage fees. It is very possible that Della Vigna and            In a similar study, Thaler and Benartzi (2004) presented
Malmendier observed this pattern of behavior because when              randomly selected employees at several different companies
thinking about the future, people naively overpredict how              with the opportunity to enroll in a savings plan called ‘‘Save
frequently they will make the should decision to go to the gym.        More Tomorrow’’ or SMarT. This savings plan invited employees
An alternative explanation, however, is that people are sophis-        to precommit to automatically placing 50% of the proceeds of
ticated about their dynamic inconsistency and sign up for un-          their future pay raises in an investment savings account. Thaler
limited gym visit contracts to increase the likelihood that their      and Benartzi found that people who were offered the opportunity
future selves will go to the gym. In other words, reducing the         to participate in this plan saved dramatically more than those
marginal cost of a gym visit to zero may serve as a commitment         who were not invited to participate. SMarT’s design capitalized
device, which increases gym attendance and thus has a positive         on several different psychological and economic principles to
value.                                                                 increase its appeal and effectiveness. For example, the plan
   In a more controlled study of commitment devices, Ariely and        offered participants the opportunity to purchase expert-recom-
Wertenbroch (2002) examined whether college students would             mended investments (reducing the cognitive costs of deciding
opt to assign themselves deadlines for three papers they were          what to invest in), invited employees to save only from future pay
required to hand in by the end of an academic semester. Stu-           increases (thus capitalizing on dynamic inconsistency and
dents who assigned themselves deadlines were committing to             preventing employees from experiencing increased savings as a
complete one or more papers before the last minute (a should           loss), and automatically deposited employees’ money in savings
behavior) rather than procrastinating for as long as possible (a       (taking advantage of the power of defaults). Because Thaler and
want behavior). The authors found that when students were gi-          Benartzi did not isolate individual features of SMarT, their re-
ven the option to assign themselves deadlines, 73% elected to          search does not reveal which specific characteristics of the plan
impose deadlines on themselves that would require them to turn         increase savings rates. However, Thaler and Benzarti’s work
in one or more of their papers before the last day of class. This      suggests that offering people the opportunity to precommit to
indicated that many students were aware of their self-control          savings products may have the potential to increase savings
problems and placed a positive value on a commitment device            rates.

Volume 3—Number 4                                                                                                                    333
You can also read